OpenCL: Compute, Cryptography, and Bandwidth
Shader Performance: FP32 vs. FP64
Let’s start with an OpenCL benchmark, which should push the theoretical ceiling of 32- and 64-bit compute performance.
Sandra's Cryptography module is next. It’s remarkable how well the older FirePro W9000 keeps up. The distance between the W9100 and the W8100 is right around where it should be based on each card's specifications.
Although these benchmarks are synthetic in nature, they still illustrate Nvidia’s half-hearted support of OpenCL. Yes, the company offers its proprietary CUDA API, and there are plenty of applications that support it. Increasingly, though, ISVs looking for a broader customer base don't want to support two languages, and OpenCL is gaining traction as a result. Even long-time bastions of CUDA support like Adobe are adopting OpenCL.
Folding It Up: Folding@Home
Let’s run the Folding@Home benchmark on this card. Even though few professionals would use a workstation-class board for this (or cryptocurrency mining), the test does give us a more real-world look at compute performance.
Once again, the gap between each card's performance is what we'd expect in light of their specifications. This chart demonstrates nicely how well AMD's architecture scales in scenarios without overhead.
Conversely, when comparing memory bandwidth under OpenCL to Direct3D 11, Nvidia demonstrates that putting more effort into optimizing drivers makes a quantifiable difference.
As we move on to our application benchmarks, keep these synthetics tests in mind. They help decipher the performance results of real-world metrics, which are subject to influence from other platform subsystems.
At least for now, we have to question whether Nvidia's lackluster support for OpenCL and emphasis on CUDA is the best strategy. Only time will tell.
Wait a minute. If this kind of cooling is better than the ones that used on R9 290 and they had this kind of technology from HD5800 series, then why in the hell they didn't use it on R9 290 series instead of using this crap cooler they used?
That's almost an understatement. The K5000 is almost constantly 50% slower than the W8100, with a few 25% cases difference. For 700 $ more, the W8100 looks like a great buy.
I think the real reason might come from your review.
I don't believe in coincidence, but they decided to use it on a more expensive professional GPU with great success.
How do we know that the cooler used in W8100 wasn't approved for R9 290(X) cause of its higher cost perhaps?
ps: Am I asking too much if I ask from any reviewer on Tom's to test this cooler on a R9 290? (if its compatible ofc...)
The focus on the Firepro W8100 and Quadro K5000 being competitors as something to directly compare is a bit misleading and distracts attention from the impressive features of the Firepro W8100.
The W8100 does outperform the Quadro K5000 is some important ways, but to be in marketing competition, the performance should be to be in the same general league. The W8100 is 56% more expensive- the price difference of $900 is more than enough to buy a K4000 (About $750).
On a marketing-basis a $60,000 car that is 50% faster is not a direct competitor to a $38,000 one. The use and expectations of performance and quality are different. The logic is to say, "If you're thinking of buying a Quadro K5000, you should know that for 56% more you can have 25-50% higher performance in several important but not all categories." These purchases are most often budget driven- how many have unlimited funds- and the buyer of a $1,600 card will be a different person from someone with a $2,500 budget. The buyer's quest is more often based on how much performance is expected combined with how much is possible within the budget.
These cards may have the same applications, but for the W8100 to be better value than a K5000, it should have a consistent 56% performance advantage. A better comparison would be to consider for example, the W7000 and Quadro K4000. Both about $750, but the W7000 is 256-bit, has 4GB. a 154GB/s bandwidth. and 1280 stream processors against the K4000's 192-bit 3GB, 134GB/s and 768 CUDA cores. On Passmark Performance Test, a W7000 3D score near but not the top is about 4300 and 2D at about 1000 while the K4000 scores near the top at about 3000 3D and 1100 2D. The news for AMD is even better when considering that a $1,600 Quadro K5000- double the W7000 cost but also 4GB and 256-bit- near the top 3D scores are about 4300 and in 2D about 900. For me, a better marketing strategy would be to compare the K5000 to the W7000 and the W8100 to a mythological "K5500" that would cost $2,800 (midway between 4 and 12GB and $1600 and $5000).
This means that the person looking for the best performance for $750 -and uses the applications the W-series is good at- has an easy choice in the W7000.
Still, the features- especially the 512-bit and 8GB plus overall performance make the W8100 one to consider in the upper end of workstation cards. This should be a very good animation /film editing card. The comments about AMD being more forward looking than NVIDIA may be correct though the comments about the quality of Quadro drivers also seems true. This furthers the trend of GPUs tending to concentrate in certain functions-( the W8100 in OpenCL for example), having to consider GPU's one by one according to the applications used. More and more, with complex 3D modeling and animation software, specific software drives graphics card choices and except for the very top of the lines, the cards seem to less all-rounders than before- not good at everything.
I assume that should be, 'clock rate'.
Btw, how come the test suite has changed so that there is no longer any
app being used such as AE for which NVIDIA cards can be strong because
of CUDA support?
present any rationale as usual. :D
And falchard is right, Viewperf tests showed enormous differences between
pro & gamer cards in previous years, but it seems vendors are deliberately
blurring the tech now, optimising for consumer APIs (ie. not OGL), which
means pro tests often run well on gamer cards. In which case where is their
rationale for the cost difference? Apart from support and supposedly better
drivers, basic performance used to be a major factor of choosing a pro card
and a sensible justification for the extra cost, but this appears to be not the
case anymore; check Viewperf11 scores for any gamer vs. pro card, the only
test where a gamer card isn't massively slower is ENSIGHT-04. For MAYA-03,
a Quadro 4000 is 3X faster than a GTX 580; for PROE-05, a Q4K is 10X faster;
for TCVIS-02, a Q4K is 30X faster.
Today though, with Viewperf12, a 580 is faster than a K5000 for MAYA-04,
about the same for CREO-01, about the same for SHOWCASE-01 and
not that much slower for SW-03. Only for CATIA-04 and SNX-02 does the
expected difference persist.
Meanwhile we get OpenCL touted everywhere, even though there are plenty
of apps which can exploit CUDA, but little attempt to properly compare the
two when the option to use the latter is also available, eg. 3DS Max, Maya,
Cinema4D, AE, LW, SI, etc.
PS. nebun, the core structure on these cards is completely different. The number
of cores is a totally useless measure, it tells one nothing. One can't even compare
between different cards from the same vendor, eg. a GTX 770 has way more cores
than a GTX 580, but a 580 hammers the 780 for CUDA. Indeed, a 580 beats all
the 600 series cards for CUDA despite having far few cores (it's because the newer
cards use a much lower core clock, less bandwidth per core, etc.)
It's nice to see that AMD is starting to close the gap on it's products. They seriously need to consider updating their cooling solutions and improving power. I would be interested to see if these workstation cards throttle down as often as their desktop counterparts. In my experience most of the current Hawaii chips are running higher voltages than needed and they could save both power and heat by running them down a bit. It should allow the boards to stay stable and compete better in many workloads.